Channel Pruning via Automatic Structure Search

Channel pruning is among the predominant approaches to compress deep neural networks. To this end, most existing pruning methods focus on selecting channels (filters) by importance/optimization or regularization based on rule-of-thumb designs, which defects in sub-optimal pruning. In this paper, we propose a new channel pruning method based on artificial bee colony algorithm (ABC), dubbed as ABCPruner, which aims to efficiently find optimal pruned structure, i.e., channel number in each layer, rather than selecting "important" channels as previous works did. To solve the intractably huge combinations of pruned structure for deep networks, we first propose to shrink the combinations where the preserved channels are limited to a specific space, thus the combinations of pruned structure can be significantly reduced. And then, we formulate the search of optimal pruned structure as an optimization problem and integrate the ABC algorithm to solve it in an automatic manner to lessen human interference. ABCPruner has been demonstrated to be more effective, which also enables the fine-tuning to be conducted efficiently in an end-to-end manner. The source codes can be available at https: //github.com/lmbxmu/ABCPruner.

LigRMSD: a web server for automatic structure matching and RMSD calculations among identical and similar compounds in protein-ligand docking

Motivation Root mean square deviation (RMSD) is one of the most useful and straightforward features for structural comparison between different conformations of the same molecule. Commonly, protein-ligand docking programs have included some utilities that allow the calculation of this value; however, they only work efficiently when exists a complete atom label equivalence between the evaluated conformations. Results We present LigRMSD, a free web-server for the automatic matching and RMSD calculations among identical or similar chemical compounds. This server allows the user to submit only a pair of identical or similar molecules or dataset of similar compounds to compare their three-dimensional conformations. Availability and implementation LigRMSD can be freely accessed at https://ligrmsd.appsbio.utalca.cl. Supplementary information Supplementary data are available at Bioinformatics online.

Automatic structure-based NMR methyl resonance assignment in large proteins

Isotopically labeled methyl groups provide NMR probes in large, otherwise deuterated proteins. However, the resonance assignment constitutes a bottleneck for broader applicability of methyl-based NMR. Here, we present the automated MethylFLYA method for the assignment of methyl groups that is based on methyl-methyl nuclear Overhauser effect spectroscopy (NOESY) peak lists. MethylFLYA is applied to five proteins (28–358 kDa) comprising a total of 708 isotope-labeled methyl groups, of which 612 contribute NOESY cross peaks. MethylFLYA confidently assigns 488 methyl groups, i.e. 80% of those with NOESY data. Of these, 459 agree with the reference, 6 were different, and 23 were without reference assignment. MethylFLYA assigns significantly more methyl groups than alternative algorithms, has an average error rate of 1%, modest runtimes of 0.4–1.2 h, and can handle arbitrary isotope labeling patterns and data from other types of NMR spectra.